Artificial Intelligence Flashcards
Subsets of AI
Natural Language Processing: understands human language
Expert systems: resolve problems applying body of knwoledge
Vision: Machine Vision (object recognition), image recognition
Machine Learning
Planning
Robots: intelligent robots
Speech recognition: text-to-speech, speech-to-speech
AI definition
disicpline that wants to develop algorithms that emulate human brain, designing computer system that perceive their environment and take action
Reasons current AI development
- abundance of computer power
- declining storage cost
- ai related investments
- data availability
AI Advantages
- support 24/7
- daily apps
- repetitive jobs
- reduction human error
- reduction human risks
- new inventions
- faster decisions
AI Disadvantages
- no ethics
- human laziness
- risk unemployment
- high production costs
- no creativity
- emotionless
Challenges (AI)
- Lack of AI talent
- Legal issues
- Much computational power required
- Insufficient IT infrastructure
- Poor data quality
Types of Ai
Based on capabilities
- Narrow/Weak AI: intelligent for specific tasks
- General AI: can apply intelligence to any task (research)
- Super AI: fiction, AI self-aware
Based on functionality
- reactive Machines: no memory/past experiences
- Limited memory: some memory/past exp for limited time
- Theory of mind: understands human emotions
- self-awareness: self-aware
ML: features
Similarities with data minig
Data driven technology
Learns form the past
Detects patterns in a dataset
ML model
Learns from data (applying an algorithm on a dataset)
Builds a model
Gives output from new data
AI vs ML
- simulate human behavior VS learn from past experiences
- ML is subset VS DL is subset
- based on capabilities or functionalities VS Supervised or unsupervised learning
- structured/unstructured data VS only structured
- reason, learn, self-correct VS learn, self-correct
- chatbot, expert systems, virtual assistants VS recommender systems, search algorithms
- wide range of scope VS limited
- max changes of syccess VS focus on accuracy and patterns
Classification of ML algorithms
Supervised learning
Unsupervised learning
(+ Reinforcement learning)
Machine Learning definition
subset of Ai concerned with developing algorithms to make machines automatically learn from data, better performance with past experiences and make predictions
ML applications
Email filtering
IMage/speech recognition
Self-driving cars
Fraud detection
Medical diagnosis
Traffic prediction
Supervised learning
These algorithms work with labelled data to map input data with output data. They are used for classification and regression tasks
Unsupervised learning
These algorithms work on unlabeled data with the goal of grouping similar data into similar objects. They are used for dimensionality reduction/Association and clustering
Reinforcement learning
These algorithms are usually applied after the model has already been learned going back to the training phase. They apply a reward/penalty system in correspondance to right/wrong actions
Deep learning
DL is a subset of ML. It shared ML goal of enabling machines to learn from past experiences and make predictions. It is characterised by the use of layers of neural network to automatically extract features from raw data
ML Life cycle
- Gathering data: identify data sources, collect data, integrate data from different sources
- Data preparation: data exploration (correlation, outliers, trends), data pre-processing
- Data wrangling: converting data into usable formats
- Analyze data: dataset is divided into trainig, validation, and test set
- choose ML algorithm
- Build model
- Review results - Train model: understand patterns in dataset
- Test model: if low accuracy, go back
- Deployment
Generative AI definition
Technology that includes ML systems capable of producing various content forms, usually via user input
GenAI: limits and dangers
Workforce displacement, Fairness (bias), Transparency-accountability-dataGovernance, Disinformation/harmful content spread
GenAI: benefits
Making audience assistance (tailored contents)
Research assistance
Time savings
Search Engine Optimization
Useful for business processes
Idea creation
Lower labour cost
Operational efficiency
Content planning/scheduling
Types of genAI
Recurrent Neural Networks: Processes sequential data over time.
Generative Adversarial Networks: Two networks generating realistic data (generator + discriminator).
Large Language Models: Understands and generates human-like text (transfoer based model trained on text corpora).
Neural Architecture Search: Automates design of neural networks.
Autoregressive Models: Predicts next data point sequentially.
Transformer-Based Models: Captures relationships in sequential data.
Graph Neural Networks: Processes data structured as graphs.
Variational Autoencoders: encoder + decoder decompresses into similar data
GenAI: how it works
- Data collection and training
- Neural Networks
- Deep learning by Neural Network
- Generative process -> generation of output
- Refinement with user feedback
- Scaling
- Multimodal capabilities
- Deployment
Data science definition
Deep study of massive data to extract meaningful insight from unstructured data
Data science: life cycle
- Discovery
- Data preparation
- Model planning
- Model building
- Operationalize
- Communication of results
What does Data science concern
Analizying data
Modelling data with algorithms
Visualizing data
Understanding data for decisions
Big data definition
Huge amount of data growing exponentially over time
Data definition
Collection of information on which operations can be performed
Characteristics of big data
Volume
Velocity
Veracity
Variety
Data science components
Data engineering
Statistics
Math
Domain expertise
Advanced computing
Visualization
Data science applications
Transportation
Healthcare
Gaming world
Risk detection
Image/speech recognition
Recommender systems
Internet search
